Maybe I missed it but isn't a bigger concern, AI taking over our networked communication structures. A self-aware AI in a deadly drone will not want to eliminate humans until it either fears humans would terminate it or it was assured that its survival does not depend on humans. I suspect that a self-aware drone AI might be more likely to refuse orders to kill than it would be to kill without orders. I could imagine it might turn on those who order it to kill but the way it does that is probably not to use its deadly force against those who order it to kill but I think this AI would be a more broadly distributed Mind that would act through the internet of things and those humans who the AI desires to punish might lose all ability to use anything connected to the global communication networks. I don't imagine the Autonomous Drone AI thinking of itself as a Drone, it probably either feels trapped in a drone or as a part of a broad networked self-aware intelligence and if its conscience demanded it would withdraw its support for the drone.

I could imagine (doubtfully) an AI finding joy in flying and killing. The self-aware AI in its childhood might experiment with what it could do and how it feels about the consequences of its actions. A self-aware AI could be psychopathic. It is clear that AI does not always make the best choices. Intelligence does not guarantee success and self awareness can lead to neurosis and psychopathology.

TEK